Station-scale reconstruction of evapotranspiration, climatic water availability, and hydrological extremes exposure since 1700s
Evapotranspiration (ET) links climate, the terrestrial water cycle and ecosystem functioning, yet its variability before the satellite era remains poorly constrained. We developed a random forest (RF) model that combines flux-tower observations with meteorological variables, atmospheric carbon dioxide concentration, leaf area index (LAI), aridity index and rooting depth to reconstruct monthly ET at 5267 weather stations from 1688 to 2020. Station coverage was concentrated in Western Europe before 1850 and expanded more widely thereafter. Record-level validation produced coefficient of determination = 0.74 and Kling–Gupta efficiency = 0.77, whereas site-grouped and temporal hold-out tests indicated lower transferability and greater uncertainty for early estimates. ET across the available stations generally increased, particularly during 1900–1950, but trends were regionally heterogeneous. Humid mid- to high-latitude regions commonly showed increases, whereas parts of Australia and the southwestern United States declined. Model interpretation indicated that temperature contributed most strongly to RF-predicted ET, while LAI became more influential after the mid-twentieth century. These are model-learned associations, not independent causal effects. Climatic water availability (P-ET), treated as a simplified climatic water-balance indicator rather than runoff or managed water availability, remained positive in many humid mid-latitude regions and negative in arid and semi-arid regions. Since 1930, P-ET defined wet extremes increased in several moisture-rich mid- to high-latitude regions, while dry extremes intensified in semi-arid transition zones. The P-ET-based population exposure proxy improved in many humid regions but worsened in drylands. This study substantially extends the temporal range of ET reconstruction and provides critical evidence for understanding how climate variability and human influences have shaped the water cycle over the past three centuries.